Commit 1167a506 authored by Cianciosa, Mark's avatar Cianciosa, Mark
Browse files

Fix issue in cuda context where kernel prefix was would wrong if an output was... 

Fix issue in cuda context where kernel prefix was would wrong if an output was reduced to a variable. Fix the jit test tolarance.
parent 7f1c51ce

graph_framework/cuda_context.hpp

+11 −11
Original line number Diff line number Diff line
@@ -756,6 +756,7 @@ namespace gpu { 
                }

            }

            for (size_t i = 0, ie = outputs.size(); i < ie; i++) {

                if (!used_args.contains(outputs[i].get())) {

                    if (i == 0) {

                        if (inputs.size()) {

                            source_buffer << ", // "
@@ -772,7 +773,6 @@ namespace gpu { 
                        source_buffer << "," << std::endl;

                    }



                if (!used_args.contains(outputs[i].get())) {

                    source_buffer << "    ";

                    jit::add_type<T> (source_buffer);

                    source_buffer << " *  __restrict__ "

graph_tests/jit_test.cpp

+1 −1
Original line number Diff line number Diff line
@@ -332,7 +332,7 @@ template<jit::float_scalar T> void run_math_tests() { 
    auto sin_node = graph::sin(v1);

    compile<T> ({graph::variable_cast(v1)},

                {sin_node}, {},

                sin_node->evaluate().at(0), 0.0);

                sin_node->evaluate().at(0), result);



    auto cos_node = graph::cos(v1);

    compile<T> ({graph::variable_cast(v1)},